Introduction

1 2 3 4 5 6

Key Questions

  • How does HIV persist and migrate across CNS and peripheral tissues despite systemic viral suppression?

Overview

This presentation summarizes:

  • Viral phylogenetics & BEAST modeling
  • CNS reservoir compartmentalization
  • Migration dynamics across tissues & BBB



Cohort Workflow

Last Gift Cohort

Body Donation

Autopsy

Tissue Collection

HIV Sequencing

Phylogenetics and Modeling

OBJECTIVES

Blood Brain Interface

Image attribution: BioRender.
(click anywhere to close)

Objectives

Aim 1

  • Identify key factors associated with viral migration across the BBB



Aim 2

  • Identify CNS-specific clusters (reservoirs)
  • Characterize:
    • Composition
    • Timing
    • Renewal from periphery
    • Reseeding capacity into peripheral reservoirs

APPROACH

Aim 1

Data Processing

  • SGA HIV env clones across tissues
  • CSF biomarkers & clinical metadata
  • BEAST diffusion + phylogenetic modeling
  • Negative Binomial & Poisson models

Workflow

Aim 2

Data Processing

  • HIV env clones across participants & tissues
  • Cleanup & alignment
  • Phylogenetic reconstruction
  • Identify CNS-specific supported clusters
  • Bayesian skyline / skygrid diffusion models
  • Timing (tMRCA) of clusters
  • Migration flows between CNS clusters & other tissues
  • Modeling factors associated with migration

Workflow

Sampling

Note

  • Our analyses included 20 participants with sequences from both the CNS and Periphery.
  • A total of 1875 SGA clones were included from 49 distincts tissues.
  • Luminex CSF data available for 34 markers.

AIM 1: Viral Dynamics

Maximum Clade Credibility Tree

Time-Scaled Maximum Clade Credibility Tree

HIV Migration Networks

HIV Flow across the BBB through time

AIM 1: Correlates of Dynamics

Overview

  • Goal: model viral migration events across the Blood Brain Barrier (CNS ↔︎ periphery).
  • Account for exposure: number of BEAST posterior states, and sequencing depth per tissue.

Data Input

Exposure

Conceptual motivation

In a transition model, exposure represents the “opportunity” for observing a migration event between compartments. We considered the following definition:

  • Exposure based on number of states weighted by product of sequence counts

Equation

\[ n_{\text{states}, i} : \text{number of Markov states in run } i \times \text{number of sequences from CNS} \times \text{number of sequences from Periphery} \]

Model Code

data %>%
  mutate(
    pairs_cns    = ntissues_cns * (ntissues_cns - 1) / 2,
    pairs_periph = ntissues_periph * (ntissues_periph - 1) / 2,
    
    exposure = case_when(
      migration_type == "cross_BBB" & direction == "CNS to Periph" ~ 
        n_states * nseq_cns * nseq_periph,
      
      migration_type == "cross_BBB" & direction == "Periphery to CNS" ~ 
        n_states * nseq_periph * nseq_cns,
      
      migration_type == "within_CNS" ~ 
        n_states * pairs_cns * (nseq_cns^2),
      
      migration_type == "within_peripheral" ~ 
        n_states * pairs_periph * (nseq_periph^2)
    )
  )

CSF Markers

Data Preparation for Migration Models

Summaries

  • Count migration/transition events from and toward the CNS
  • Compute number of discrete viral states per run
  • Count SGA clones sampled in CNS and in Periphery

Exposure Definition

  • Aggregated: # states × # CNS sequences × # Periphery sequences
  • long format: # sequences from × # sequences to

Filtering

  • Exclude virally suppressed individuals
  • Keep directional transitions:
    • CNS → Periphery
    • Periphery → CNS
    • Bidirectional

Data Transformations

  • Log-transform exposure
  • Log1p-transform CSF marker values

Workflow

Model Specification: Negative Binomial GLMM

Notes

  • Outcome: count of CNS-related migration/transition events
  • Offset: log(exposure)
  • Predictor: log1p-transformed CSF marker
  • Random effect for participants

Equation

\[ \log(\mu_i) = \beta_0 + \beta_1 \cdot \text{Marker}_i + b_{\text{pid}(i)} + \log(\text{Exposure}_i) \]

Model Code

fit_models <- function(marker, data_in) {
  formula <- as.formula(
    paste0(
      "n_events ~ ", marker,
      " + age + sex + last_cd4_t_cell_count + duration_infection_years +
        (1 | pid) + offset(log_exposure)"
    )
  )

  glmmTMB(
    formula,
    data = data_in,
    family = nbinom2(),
    control = glmmTMBControl(
      optimizer = optim,
      optArgs = list(method = "BFGS", maxit = 5000)
    )
  )
}

Model Results (aggregated)

Negative Binomial Models
Associations between CSF biomarkers and counts of migration events from the CNS to the periphery
Marker Samples Detectable1 Effect Status exp(Beta) 95% CI p-value Significance Direction
Eotaxin (CCL11) 13 Ok 1.035 0.7967–1.35 0.795
GM-CSF 8 Ok 0.884 0.7446–1.05 0.161
GRO-alpha (CXCL1) 13 Ok 0.988 0.9808–1.00 0.001 **
IL-1α 1 Sparse Data 0.160 0.0022–11.58 0.401
IL-1β 5 Ok 0.806 0.6734–0.96 0.018 *
IL-1RA 12 Ok 1.000 0.9995–1.00 0.000 ***
IL-2 2 Sparse Data 0.418 0.2951–0.59 0.000 ***
IL-5 5 Ok 0.985 0.8082–1.20 0.885
IL-6 12 Ok 0.998 0.9971–1.00 0.000 ***
IL-7 13 Ok 0.356 0.0618–2.05 0.248
IL-8 (CXCL8) 13 Ok 0.998 0.9952–1.00 0.043 *
IL-9 1 Sparse Data 1.104 0.5813–2.10 0.762
IL-10 8 Ok 0.481 0.2551–0.91 0.023 *
IL-15 5 Ok 0.442 0.1617–1.21 0.112
IL-17A/CTLA-8 1 Sparse Data 0.409 0.0534–3.13 0.389
IL-18 13 Ok 1.009 0.9871–1.03 0.435
IL-31 1 Sparse Data 0.932 0.6748–1.29 0.668
IP-10 (CXCL10) 13 Ok 1.000 0.9967–1.00 0.841
MCP-1 (CCL2) 13 Ok 1.000 0.9985–1.00 0.908
MIP-1α (CCL3) 13 Ok 0.868 0.7918–0.95 0.002 **
MIP-1β (CCL4) 13 Ok 0.964 0.9420–0.99 0.002 **
RANTES (CCL5) 13 Ok 1.007 0.9176–1.11 0.881
SDF-1α 13 Ok 1.000 0.9993–1.00 0.798
TNF-α 2 Sparse Data 0.328 0.2148–0.50 0.000 ***
1 Number of samples with ≥1 measure(s) above the LOD across the 3 replicates.
Negative Binomial Models
Associations between CSF biomarkers and counts of migration events from the periphery to the CNS
Marker Samples Detectable1 Effect Status exp(Beta) 95% CI p-value Significance Direction
Eotaxin (CCL11) 16 Ok 1.002 0.875–1.15 0.972
GM-CSF 11 Ok 0.849 0.752–0.96 0.008 **
GRO-alpha (CXCL1) 16 Ok 0.987 0.980–0.99 0.000 ***
IL-1α 1 Sparse Data 0.258 0.011–6.02 0.399
IL-1β 7 Ok 0.843 0.721–0.99 0.034 *
IL-1RA 15 Ok 1.000 1.000–1.00 0.007 **
IL-2 2 Sparse Data 0.374 0.283–0.49 0.000 ***
IL-5 7 Ok 0.893 0.678–1.18 0.420
IL-6 15 Ok 0.998 0.997–1.00 0.000 ***
IL-7 16 Ok 1.460 0.378–5.64 0.583
IL-8 (CXCL8) 16 Ok 0.999 0.998–1.00 0.140
IL-9 1 Sparse Data 1.331 0.731–2.42 0.349
IL-10 10 Ok 0.823 0.685–0.99 0.038 *
IL-15 7 Ok 0.249 0.119–0.52 0.000 ***
IL-17A/CTLA-8 2 Sparse Data 0.745 0.250–2.22 0.596
IL-18 16 Ok 1.008 0.991–1.03 0.359
IL-31 1 Sparse Data 0.911 0.689–1.20 0.509
IP-10 (CXCL10) 16 Ok 1.000 0.999–1.00 0.746
MCP-1 (CCL2) 16 Ok 1.000 0.999–1.00 0.987
MIP-1α (CCL3) 16 Ok 0.903 0.827–0.99 0.024 *
MIP-1β (CCL4) 16 Ok 0.987 0.969–1.01 0.157
RANTES (CCL5) 16 Ok 1.001 0.935–1.07 0.987
SDF-1α 16 Ok 1.000 0.999–1.00 0.296
TNF-α 3 Sparse Data 0.380 0.220–0.66 0.001 ***
1 Number of samples with ≥1 measure(s) above the LOD across the 3 replicates.
Negative Binomial Models
Associations between CSF biomarkers and counts of migration events across the BBB
Marker Samples Detectable1 Effect Status exp(Beta) 95% CI p-value Significance Direction
Eotaxin (CCL11) 16 Ok 1.079 0.9085–1.28 0.387
GM-CSF 11 Ok 0.853 0.7522–0.97 0.013 *
GRO-alpha (CXCL1) 16 Ok 0.989 0.9803–1.00 0.007 **
IL-1α 1 Sparse Data 0.295 0.0061–14.32 0.538
IL-1β 7 Ok 0.824 0.6595–1.03 0.089 .
IL-1RA 15 Ok 1.000 0.9995–1.00 0.000 ***
IL-2 2 Sparse Data 0.385 0.3005–0.49 0.000 ***
IL-5 7 Ok 0.913 0.7685–1.08 0.300
IL-6 15 Ok 0.998 0.9973–1.00 0.000 ***
IL-7 16 Ok 1.273 0.2751–5.89 0.757
IL-8 (CXCL8) 16 Ok 0.999 0.9970–1.00 0.241
IL-9 1 Sparse Data 1.235 0.7163–2.13 0.448
IL-10 10 Ok 0.794 0.6189–1.02 0.069 .
IL-15 7 Ok 0.286 0.1314–0.62 0.002 **
IL-17A/CTLA-8 2 Sparse Data 1.011 0.2677–3.82 0.987
IL-18 16 Ok 1.010 0.9911–1.03 0.313
IL-31 1 Sparse Data 0.955 0.7104–1.28 0.762
IP-10 (CXCL10) 16 Ok 1.000 0.9988–1.00 0.576
MCP-1 (CCL2) 16 Ok 1.000 0.9987–1.00 0.763
MIP-1α (CCL3) 16 Ok 0.883 0.7876–0.99 0.032 *
MIP-1β (CCL4) 16 Ok 0.988 0.9626–1.01 0.343
RANTES (CCL5) 16 Ok 1.024 0.9410–1.11 0.581
SDF-1α 16 Ok 1.000 0.9976–1.00 0.996
TNF-α 3 Sparse Data 0.358 0.2383–0.54 0.000 ***
1 Number of samples with ≥1 measure(s) above the LOD across the 3 replicates.

Model Results (long)

Negative Binomial Models (per transition)
Associations between CSF biomarkers and counts of migration events from the CNS to the periphery
Marker Samples Detectable1 Effect Status exp(Beta) 95% CI p-value Significance Direction
Eotaxin (CCL11) 13 Ok 1.060 0.930–1.21 0.385
GM-CSF 8 Ok 0.939 0.863–1.02 0.140
GRO-alpha (CXCL1) 13 Ok 0.996 0.992–1.00 0.090 .
IL-1α 1 Sparse Data 0.248 0.032–1.92 0.181
IL-1β 5 Ok 0.986 0.856–1.13 0.840
IL-1RA 12 Ok 1.000 1.000–1.00 0.001 ***
IL-2 2 Sparse Data 0.668 0.515–0.87 0.002 **
IL-5 5 Ok 0.997 0.904–1.10 0.954
IL-6 12 Ok 0.999 0.999–1.00 0.068 .
IL-7 13 Ok 1.570 0.581–4.24 0.373
IL-8 (CXCL8) 13 Ok 0.999 0.998–1.00 0.033 *
IL-9 1 Sparse Data 0.916 0.662–1.27 0.595
IL-10 8 Ok 0.648 0.478–0.88 0.005 **
IL-15 5 Ok 0.666 0.398–1.11 0.121
IL-17A/CTLA-8 1 Sparse Data 0.555 0.203–1.52 0.252
IL-18 13 Ok 1.009 0.998–1.02 0.098 .
IL-31 1 Sparse Data 1.044 0.886–1.23 0.609
IP-10 (CXCL10) 13 Ok 1.000 0.999–1.00 0.832
MCP-1 (CCL2) 13 Ok 1.000 0.999–1.00 0.451
MIP-1α (CCL3) 13 Ok 0.963 0.892–1.04 0.340
MIP-1β (CCL4) 13 Ok 0.985 0.969–1.00 0.067 .
RANTES (CCL5) 13 Ok 1.009 0.963–1.06 0.699
SDF-1α 13 Ok 1.000 1.000–1.00 0.897
TNF-α 2 Sparse Data 0.593 0.436–0.81 0.001 ***
1 Number of samples with ≥1 measure(s) above the LOD across the 3 replicates.
Negative Binomial Models (per transition)
Associations between CSF biomarkers and counts of migration events from the periphery to the CNS
Marker Samples Detectable1 Effect Status exp(Beta) 95% CI p-value Significance Direction
Eotaxin (CCL11) 16 Ok 1.017 0.927–1.12 0.716
GM-CSF 11 Ok 0.910 0.862–0.96 0.001 ***
GRO-alpha (CXCL1) 16 Ok 0.995 0.991–1.00 0.002 **
IL-1α 1 Sparse Data 0.622 0.085–4.52 0.639
IL-1β 7 Ok 0.998 0.879–1.13 0.970
IL-1RA 15 Ok 1.000 1.000–1.00 0.001 **
IL-2 2 Sparse Data 0.645 0.539–0.77 0.000 ***
IL-5 7 Ok 0.963 0.886–1.05 0.376
IL-6 15 Ok 0.999 0.999–1.00 0.003 **
IL-7 16 Ok 1.645 0.746–3.63 0.217
IL-8 (CXCL8) 16 Ok 0.999 0.998–1.00 0.109
IL-9 1 Sparse Data 1.007 0.761–1.33 0.960
IL-10 10 Ok 0.862 0.748–0.99 0.042 *
IL-15 7 Ok 0.530 0.378–0.74 0.000 ***
IL-17A/CTLA-8 2 Sparse Data 0.872 0.436–1.74 0.697
IL-18 16 Ok 1.012 1.004–1.02 0.005 **
IL-31 1 Sparse Data 1.094 0.950–1.26 0.213
IP-10 (CXCL10) 16 Ok 1.000 0.999–1.00 0.973
MCP-1 (CCL2) 16 Ok 1.000 0.999–1.00 0.627
MIP-1α (CCL3) 16 Ok 0.978 0.911–1.05 0.532
MIP-1β (CCL4) 16 Ok 0.993 0.979–1.01 0.291
RANTES (CCL5) 16 Ok 1.001 0.961–1.04 0.945
SDF-1α 16 Ok 1.000 0.999–1.00 0.885
TNF-α 3 Sparse Data 0.573 0.458–0.72 0.000 ***
1 Number of samples with ≥1 measure(s) above the LOD across the 3 replicates.
Negative Binomial Models (per transition)
Associations between CSF biomarkers and counts of migration events across the BBB
Marker Samples Detectable1 Effect Status exp(Beta) 95% CI p-value Significance Direction
Eotaxin (CCL11) 16 Ok 1.022 0.933–1.12 0.638
GM-CSF 11 Ok 0.917 0.864–0.97 0.004 **
GRO-alpha (CXCL1) 16 Ok 0.995 0.991–1.00 0.004 **
IL-1α 1 Sparse Data 0.418 0.062–2.82 0.370
IL-1β 7 Ok 0.983 0.867–1.11 0.782
IL-1RA 15 Ok 1.000 1.000–1.00 0.000 ***
IL-2 2 Sparse Data 0.643 0.539–0.77 0.000 ***
IL-5 7 Ok 0.966 0.889–1.05 0.411
IL-6 15 Ok 0.999 0.999–1.00 0.003 **
IL-7 16 Ok 1.580 0.722–3.46 0.252
IL-8 (CXCL8) 16 Ok 0.999 0.998–1.00 0.050 *
IL-9 1 Sparse Data 0.980 0.744–1.29 0.884
IL-10 10 Ok 0.867 0.752–1.00 0.048 *
IL-15 7 Ok 0.543 0.379–0.78 0.001 ***
IL-17A/CTLA-8 2 Sparse Data 0.863 0.432–1.72 0.677
IL-18 16 Ok 1.011 1.003–1.02 0.005 **
IL-31 1 Sparse Data 1.074 0.931–1.24 0.329
IP-10 (CXCL10) 16 Ok 1.000 0.999–1.00 0.965
MCP-1 (CCL2) 16 Ok 1.000 0.999–1.00 0.482
MIP-1α (CCL3) 16 Ok 0.968 0.904–1.04 0.350
MIP-1β (CCL4) 16 Ok 0.991 0.978–1.00 0.206
RANTES (CCL5) 16 Ok 1.005 0.964–1.05 0.810
SDF-1α 16 Ok 1.000 1.000–1.00 0.953
TNF-α 3 Sparse Data 0.576 0.464–0.72 0.000 ***
1 Number of samples with ≥1 measure(s) above the LOD across the 3 replicates.

Thank you